Has the cystal ball forseeing future Pub Prez primary percentages been invented?

[Torie]

One thing that is interesting about the Illinois PPP poll, is that its number for Mittens (after adding in another percent or two from the 3% undecided), just about exactly matches the 47% Mittens percentage which Sean Trende projects using his most sophisticated regression analysis formula, generated by county level (1) Evangelical (Rick), (2) black (Rick for very complex reasons, even though it should be pro Romney (if all the Evangelicals are blacks who don't vote in the GOP primary that doesn't help Rick)), (3) education level (Mittens), and (4) LDS, percentages (Mittens!).  I have labeled which variable is pro which candidate.  The Evangelical variable is the most salient. Who knew?

As to education level, that could be a proxy for an anti-Evangelical more secular vote (as opposed to non-Evangelicals who are not anti-Evangelical fundamentalists), or it could be as Trende suspects, is really a proxy for wealth, and better than wealth because in absolute numbers wealth varies by region, and adjusted for cost of living, and relative to your community, the same types of folks adjusted for the other variables may make $150,000 plus in NYC, and $75,000 plus in Ft. Smith, Arkansas. Or maybe being smarter means you are more likely to vote for Mittens, because he's just objectively the best.

Anyway, below is the projection chart. Just use the Trende percentages. The Enten percentages are close to GIGO, because he just used state level data, rather than county data, and thus has only 5 data points.  Using just 5 statewide past results "regressioned" against the four variables, is a  dog that just won't hunt; you need more data points than that (which Trende gets by instead of using just five statewide results, using as many data point results as the 5 states have counties) to generate an acceptable T stat and confidence level.

So not only with PPP be tested tomorrow, but so will Trende's regression equation.

[Torie]

It is odd this thread got no interest. I thought it one of my best efforts. I'm hurt!  Don't you guys like statistics? 

[🐒Gods of Prosperity🔱🐲💸]

Oh, it's fascinating. I just didn't have anything to add.

[Bacon King]

Interesting analysis tool, if a bit simplistic. It seems that it doesn't directly predict outcomes, actually, but the Romney and anti-Romney shares of the non-Paul vote?

On that note, by the way, the model isn't predicting that Romney gets 47% of the vote, but 47% of the three-way race with Paul excluded. Assuming PPP is right and Paul gets 10% (probably an overestimate, but still) that'll give Romney 42% of the overall vote, fairly below what that PPP poll has him at (even before you take the undecideds into account).

I'm pretty sure it'd actually be fairly simple to find a demographic correlation to Ron Paul's vote, by the way (once adjusted for caucus/primary states, and open/closed contests, of course)- certainly a strong relationship with the number of college students in a county, for example.

But back to the main topic: the correlation with white non-Mormon evangelicals voting for the Anti-Romney is obvious, but besides that I think this analysis is too simplistic to really be incredibly accurate. The correlation between education and Romney support doesn't hold when the educateds are all liberals (see: Alachua County, FL, for example). I'm also assuming that the link with Hispanics and Romney is mostly coincidental, and just based on Romney getting the support of the Cuban establishment in Miami and then winning NV without a real contest.

[Torie]

Good point about the 3 way vote, which I missed.  As to educateds, in most places, even where most are liberals, the GOP vote (even if relatively low) is probably still more educated than in counties with relatively few educateds.  Anyway, the regression held up quite well. Sure some counties will be outliers. Hispanics were not a variable in the regression equation.

[Bacon King]

Good point about the 3 way vote, which I missed.  As to educateds, in most places, even where most are liberals, the GOP vote (even if relatively low) is probably still more educated than in counties with relatively few educateds.  Anyway, the regression held up quite well. Sure some counties will be outliers. Hispanics were not a variable in the regression equation.

Fair point regarding the educateds; I suppose that would really be more of a problem when you're trying to figure out relative turnout between counties to combine everything into a statewide result. I wonder how they do estimate state results based on these county regressions, actually; I haven't found it mentioned anywhere.

Regarding Hispanic population, it seems that it wasn't considered as a variable at first, but it was later added in. See this page, towards the bottom, as well as the top of the next page:

You must be logged in to read this quote.

And yes, I agree that it's held up fairly well, which is pretty impressive, but at the same time all it's really doing is telling us that Mormons and the upper class will vote Romney, while poors, evangelicals, and whites in the black belt generally won't. Pretty much anyone here already knows that

What I'd really like to see, though, is some sort of demographic correlation that can determine the anti-Romney split between Gingrich and Santorum. No idea how you could do that, besides maybe looking at something like "American ethnicity" as a pro-Gingrich factor?

[Torie]

You could run a regression on the Gingrich vote, and see how it compares to the Mittens demographic. But you need to do it only with states post Florida where Gingrich is down below 20%, where he is now. The theory is that what Newt lost so far, are mostly now Rick voters, while what remains would be more evenly split, which makes sense, since they are  presumably relatively more indifferent between Rick and Mitt, and less likely to be anyone but Mittens voters.

As to the Hispanic thing, in states where Mittens has run relatively poorly, there are fewer Hispanics. I suspect that variable is noise. The blacks had to be included, to get a feel for the size of the white evangelical vote. It is odd that black variable still was anti Mittens (it should have been pro Mittens, to offset the evangelical variable, where a big hunk of the evangelicals are black), which means that after correcting for everything else, having blacks around eroded the Mittens vote. Maybe the most racist are less pro-Mittens, and responded to the food stamp President Newt chat or something. That aspect was not adequately discussed in the presentation of the regression equation.

[TheDeadFlagBlues]

There might be a few exceptions to the rule: non-Evangelicals in Louisiana have the potential to go as strongly, if not more strongly, for Santorum as Evangelicals. Cajuns are a completely different ball game. Also: it's quite possible that Trende is overestimating how strongly downscale non-Evangelicals vote for Romney in the midwest. Have we really had large sample sizes from rural counties that are Protestant or Catholic?

[Torie]

There might be a few exceptions to the rule: non-Evangelicals in Louisiana have the potential to go as strongly, if not more strongly, for Santorum as Evangelicals. Cajuns are a completely different ball game. Also: it's quite possible that Trende is overestimating how strongly downscale non-Evangelicals vote for Romney in the midwest. Have we really had large sample sizes from rural counties that are Protestant or Catholic?

Yes, there are a lot of counties, and some in the mix are rural Catholic counties. That is the beauty of the regression analysis.